Comment on: Risk factors for severe cranial ischaemic events in an Italian population-based cohort of patients with giant cell arteritis.

SIR, Salvarani et al. [1] recently published an article on risk factors for severe cranial ischaemic events (CIEs) in GCA. Some words of caution are warranted regarding the results of the study and their interpretation. First, 33/38 included patients (87%) with CIE had visual loss, mostly due to anterior ischaemic optic neuritis. These patients largely outweigh those with cerebrovascular accidents and whether the results apply to the latter is unknown. Hence, this work should be viewed as a study on risk factors for visual loss, rather than CIE, in GCA. Secondly, it is especially worrying to read a claim of causation in the second key message, stating that ‘anti-platelet or anticoagulation therapy did not reduce the risk of cranial ischaemic events in GCA’. The positive association between anti-thrombotic therapy and CIE probably reveals the unsurprising fact that antecedent diseases treated with anti-thrombotic agents are associated with an increased risk of thrombotic events. Likewise, the conclusive causal statement that ‘a lower inflammatory state appears to contribute to the development of severe CIE in GCA’ is premature. Elderly patients without visual loss need cranial symptoms, systemic manifestations or unexplained inflammation in order to suspect GCA. It was therefore expected that these features would be more prevalent in the group of patients without visual loss at diagnosis. Thirdly, the discussion and the first key message focus on the results of the multivariable analysis, which should be interpreted very cautiously. The multivariable model was built on 120/180 patients at most, because CRP levels were available for only 16/38 patients (42%) with CIE and 104/142 patients (73%) without CIE. We are not informed if patients with CRP results were similar to those without CRP results and why much fewer CRP results were available for patients with CIE than for those without. Moreover, with only 16 patients available in the CIE group and four predictor variables, this logistic regression model is far from having the required 10 outcome events for each predictor variable [2]. These issues threaten the validity of the multivariable analysis. Even if these methodological concerns are ignored, the authors should have concluded that no variable was associated with CIE in multivariate analysis since all P-values are >0.05. The lack of statistical significance should have precluded any discussion on the clinical meaning of these results.